疟疾与气象因素关系不同模型预测效果比较

发布时间：2018-02-21 14:30

本文关键词： 疟疾气象因素 BP神经网络模型逐步回归模型预测效果　出处：《中国公共卫生》2017年06期 　论文类型：期刊论文

【摘要】：目的分析气象因素与海南省万宁市疟疾发病率的相关性,比较BP神经网络模型和逐步回归模型对疟疾发病率的预测效果。方法收集1995年1月—2007年12月万宁市每月气象数据和疟疾发病率数据,应用Spearman等级相关分析方法分析气象因素与疟疾发病率之间的相关性,分别用BP人工神经网络方法和逐步回归方法建立疟疾发病率的气象因子拟合模型,预测2008年各月的疟疾发病率。结果万宁市疟疾月发病率与前1个月的平均气温、最高气温、最低气温、降雨量、日照时间均呈正相关(均P0.05),与前1个月的平均相对湿度、平均气压均呈负相关(均P0.01);将7种气象因素作为输入变量,疟疾发病率作为输出变量,构建内含1个隐含层的BP神经网络模型,在隐单元数为16时拟合效果最优,经过300次训练达到设定的最小训练误差为0.001,模型的均方误差和决定系数R~2分别为0.002 7和0.99;将7种气象因素作为自变量,疟疾发病率作为因变量构建逐步回归模型,进入模型的变量为平均气温和平均相对湿度,模型的决定系数R~2为0.40;应用2种模型对2008年各月疟疾发病率进行预测,平均绝对误差分别为1.24/10 000和0.44/10 000。结论万宁市疟疾发病率与气象因素明显相关,利用气象因素构建的BP神经网络模型较逐步回归模型具有更好的发病率拟合效果,但逐步回归模型的预测效果更好,BP神经网络模型的泛化能力需要进一步提高。
[Abstract]:Objective to analyze the correlation between meteorological factors and malaria incidence in Wanning, Hainan Province. The prediction effect of BP neural network model and stepwise regression model on malaria incidence was compared. Methods monthly meteorological data and malaria incidence data were collected from January 1995 to December 2007 in Wanning city. The correlation between meteorological factors and malaria incidence was analyzed by using Spearman grade correlation analysis method. The fitting models of meteorological factors were established by BP artificial neural network method and stepwise regression method respectively. Results the monthly incidence of malaria in Wanning was positively correlated with the average temperature, maximum temperature, minimum temperature, rainfall and sunshine time of the previous month (all P0.05%, and the average relative humidity of the previous month), the incidence rate of malaria in Wanning city was positively correlated with the average temperature, maximum temperature, minimum temperature, rainfall and sunshine time of the previous month. The mean pressure was negatively correlated (P0.01C), 7 meteorological factors were taken as input variables and malaria incidence rate was used as output variable, BP neural network model with one hidden layer was constructed, and the best fitting effect was obtained when the number of hidden units was 16:00. After 300 times of training, the minimum training error is 0.001, the mean square error and determination coefficient Rn2 of the model are 0.0027 and 0.99.The seven meteorological factors are taken as independent variables and the malaria incidence rate is used as dependent variable to construct a stepwise regression model. The variables entering the model were mean air temperature and mean relative humidity, and the determination coefficient Rn2 of the model was 0.40.The incidence of malaria in 2008 was predicted by two kinds of models. The average absolute error was 1.24 / 10 000 and 0.44 / 10 000 respectively. Conclusion the incidence of malaria in Wanning is significantly related to meteorological factors. The BP neural network model based on meteorological factors has better fitting effect than stepwise regression model. But the prediction effect of stepwise regression model is better than that of BP neural network model.
【作者单位】：军事医学科学院疾病预防控制所全军疾病监测中心;海南省万宁市卫生防疫站;
【基金】：国家自然科学基金(30872151)
【分类号】：R531.3

【参考文献】